Week 03 — Classification — Logistic Regression, k-NN, Naive Bayes

Fisher's 1936 iris dataset: the first formal classification algorithm. The descendants now classify spam, fraud, tumors, signals.

Lecture

Logistic regression as MLE under Bernoulli noise · linear and quadratic discriminant analysis · $k$-NN and the curse of dimensionality · Naive Bayes and the Sahami 1998 spam filter.

Read before the lecture

• Hastie, Tibshirani, Friedman, chapter 4

Code lab

Lab 1 — Classification on a clinical dataset

Build a binary classifier predicting 30-day hospital readmission. Compare logistic regression, $k$-NN, Naive Bayes, and a calibration-aware variant.

Notebook: lab01-clinical-classification.ipynb  ·  Dataset: MIMIC-IV demo (1,000 patients, redistributed).

Reference text for this week: chapter 03 of the bilingual notes — EN PDF · FR PDF.